CALL FOR PAPERS
Special Topic on Learning in Large Probabilistic Environments
Shie Mannor and
We invite papers on learning in large probabilistic environments
for a special topic of the Journal of Machine Learning Research
(JMLR). One of the fundamental problems of artificial
Intelligence is how to enable systems (for example, mobile
robots, manufacturing systems, or diagnostic systems) embedded in
complex environments to achieve their long-term goals
efficiently. A natural approach is to model such systems as
agents that interact with their environment through actions,
perceptions and rewards. These agents choose actions after every
observation, aiming to maximize their long-term reward. Learning
allows them to improve their initial strategy based on the
history of successful and unsuccessful interactions with the
This special topic is intended to serve as an outlet for
recent advances in learning in such environments, often
called reinforcement learning. We welcome both theoretical
advances in this field as well as detailed reports on
applications of learning in large probabilistic domains.
Topics of interest include:
- Theoretical foundations of learning in large probabilistic
- Completely and partially observable Markov decision process
models (MDPs) and similar models. Learning with factored state or
action spaces, continuous state spaces, action spaces or time
models, hybrid models, relational learning, concurrency.
- Heuristics and approximations. Policy and value function
approximations, Monte Carlo and advanced simulation methods.
- Spatio-temporal abstractions. Dynamic factorization, hierarchy
and relational structure.
- Interactive learning. Guided exploration, combining supervised
and unsupervised learning, shaping, and learning from very few
- Learning in complex systems. Function approximation,
dimensionality reduction, feature selection for learning, and
alternative state representations.
- Cooperative and competitive multi-agent reinforcement
learning. Learning in nonstationary domains and stochastic,
network, and dynamic games.
- Real world applications. Medicine, finance, robotics,
manufacturing, security, etc.
Submit papers to the standard JMLR submission system
Please include a note stating that your submission is for
the special topic on Learning in Large Probabilistic
Environments. Accepted papers will be published in JMLR as
they become available.
For further details or enquiries, please contact the guest
- Submission due: June 1st, 2005
- Decision: August 1st, 2005
- Final version due: October 1st, 2005
Sven Koenig (firstname.lastname@example.org)
Shie Mannor (email@example.com)
Georgios Theocharous (firstname.lastname@example.org)